Evaluation of time series models for influenza trends accross climatic zones in Cameroon (2017-2022)

doi:10.21203/rs.3.rs-7887920/v1

Download PDF

Research Article

Evaluation of time series models for influenza trends accross climatic zones in Cameroon (2017-2022)

https://doi.org/10.21203/rs.3.rs-7887920/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background:

Influenza seasonality is well established in temperate regions, where activity typically peaks in winter. In tropical countries, however, seasonal patterns remain less clear. This study compared two time series models to characterize influenza trends in four climatically distinct regions of Cameroon.

Methods:

The study covered January 2017 to December 2022. Patients with acute respiratory infections were enrolled through sentinel influenza surveillance sites in the Centre, Littoral, North, and West regions. Samples were tested for influenza virus using molecular assays. Two models were evaluated: (1) an ARIMA model incorporating 12 climatic variables (temperature, humidity, precipitation, solar radiation) and (2) a seasonal model excluding climate inputs. Model performance was assessed using root mean square error (RMSE) and stationary R².

Results:

Overall, the seasonal model outperformed the ARIMA model, with explanatory power ranging from 66% to 82%. The ARIMA model showed variable performance by region and was influenced by specific climatic factors. In the Centre and North regions, precipitation explained about 30% of influenza activity, while in the Littoral and West, temperature accounted for up to 30.8% and 43.5%, respectively. Humidity and solar radiation showed weaker associations. In the Centre, the seasonal model achieved a lower RMSE (16.067) and higher stationary R² (82%) compared with ARIMA (RMSE = 20.203; R² ~19%). Similar trends were observed across the other regions.

Conclusions:

Findings highlight heterogeneity in influenza patterns across Cameroon. Seasonal fluctuations in influenza activity appear to be shaped by factors beyond climatological variables alone.

Influenza

ARIMA model

Time-series analysis

Seasonal pattern

Disease dynamics

Cameroon

Influenza remains a major global public health challenge. Each year, an estimated 1 billion cases occur worldwide, including 3 to 5 million severe cases and 290,000 to 650,000 influenza-related respiratory deaths.[1] The disease burden is disproportionately high in resource-limited countries, particularly those in tropical regions,[2] where influenza also imposes a significant economic strain on healthcare systems and society. Influenza viruses belong to the Orthomyxoviridae family and possess a segmented, single-stranded, negative-sense RNA genome. Seasonal influenza is primarily caused by influenza A and B viruses, which circulate globally in various subtypes.[3] However, the relationship between weather conditions and influenza activity varies across regions. In Cameroon, seasonal transmission patterns are heterogenous, even among neighbouring areas with similar climatic profiles.

In temperate regions, influenza seasonality is well characterized with activity typically peaking during the winter months.[4–6] In contrast, tropical regions experience influenza circulation year-round, with incidence often linked to climatic variables that vary by location.[7–11] For example, low rainfall and temperature have been associated with increased influenza activity in two cities in Pakistan,[12] while in Lanzhou, China, a sub-tropical zone, low ambient temperature, low humidity, and high air pressure were identified as risk factors.[13] Rainfall has been shown to partially explain influenza trends in Côte d’Ivoire and Cameroon[10, 14], whereas specific humidity was the dominant driver in other tropical settings.[8, 15]

Influenza surveillance data from Cameroon, as reported in the study by Monamele et al.,[16] indicated detection rates of 24.4% prior to the COVID-19 pandemic and 15% during the pandemic period, reflecting substantial viral circulation and temporal variability. Despite these findings, the seasonality of influenza remains poorly understood, and the influence of climatological factors on its circulation has yet to be clearly defined. A study by Njifon et al. (2018) found that a one-unit increase in humidity was associated with a more than 65% rise in influenza activity two months later in arid zones of Cameroon’s Northern region.¹¹ Similarly, Monamele et al. (2017) reported weak associations between influenza activity and standard meteorological parameters across five sentinel sites in Yaoundé.[10] They hypothesized that unmeasured factors such as ultraviolet radiation and absolute humidity might better explain observed influenza peaks. This study incorporates additional climatic variables previously overlooked in prior studies from Cameroon. Given the apparent seasonal trends in influenza activity, we hypothesize that a seasonal time series model may better capture influenza dynamics than traditional climate-driven models. To test this hypothesis, we compared two models for describing and predicting influenza activity across four distinct climatic zones in Cameroon. The first model integrates twelve climatological variables, including temperature, humidity, precipitation and solar radiation, while the second is a simple seasonal model. A deeper understanding of the factors driving influenza seasonality could enhance forecasting, inform vaccination strategies, strengthen health care preparedness, and ultimately reduce the disease burden in affected communities.

Data sources and study design

This cross-sectional time-series study was conducted between January 2017 and December 2022. Thirteen influenza sentinel surveillance sites were selected across four regions of Cameroon (Centre, Littoral, West, and North), which historically have more regular and complete influenza data than the sentinels sites located in the country’s six other regions. Most positive cases were suspected influenza infections, confirmed by RT-PCR. Clinical inclusion criteria followed the World Health Organization (WHO) case definition for influenza-like illness (ILI) and severe acute respiratory infection (SARI): patients of all ages presenting with sudden onset of fever (≥ 38°C), cough and/or sore throat within the previous ten days were systematically recruited. Among the thirteen sites, four received both inpatients and outpatients: "Centre Hospitalier d'Essos" and "Hospital Jamot" (Centre region), "Hôpital Catholique Saint Albert le Grand” (Littoral region), and Garoua Regional Hospital (North region). The remaining nine sites received only outpatients. These are : "Centre Médico-Social Ambassade de France", "Centre d'Animation Sociale et Sanitaire de Nkolndongo" and "Centre Médical d'Etoudi" (Centre region); "Hôpital Catholique Notre dame de Logpom" (Littoral region); "Centre de Santé Médicalisé de Semto-Bandjoun" and "Centre de Santé Médicalisé de Kueka - Foumban" (West region); "Centre de Santé Intégré de Foulbere", "Centre de Santé Intégré de Poumpouré", and "Centre de Santé Intégré de Roumde Adja" in the North region. Figure 1 shows the location of the study sites.

Geographic and climatic context

Cameroon covers a surface area of 475,650 km², and had an estimated population of 27.6 million in 2022. It is bordered by Lake Chad to the north, the Republic of Chad to the northeast, the Central African Republic to the east, the Republic of Congo, Gabon and Equatorial Guinea to the south, and Nigeria to the west. Cameroon lies at approximately 7.3697° N latitude and 12.35° E longitude, placing it in the northern and eastern hemispheres above the equator.

According to the Köppen-Geiger classification,[17]Cameroon experiences a primary rainy season from May to November, driven by the West African Monsoon. Peak rainy months coincide with the lowest average temperatures. The Southern Plateaus have two shorter rainy seasons: May to June and October to November. The dry season spans December to April, with the highest temperatures typically recorded between February and April. Southern Cameroon is humid and equatorial with temperatures ranging from 20–25°C (depending on altitude), and monthly rainfall exceeding 400 mm in the wettest areas. In contrast, northern Cameroon (north of 6° latitude) is semi-arid, with temperatures ranging from 25–30°C and monthly rainfall below 100 mm.[17]

Laboratory analyses

Viral RNA was extracted from the nasopharyngeal swabs using the QIamp viral RNA Mini kit (Qiagen®, Hilden, Germany), following the manufacturer's procedure. Influenza virus detection was performed using the CDC Influenza Typing Assay on an ABI Prism 7500 thermocycler (Applied Biosystem, Foster City, CA, USA), considered the gold standard for influenza diagnosis.

A 20µL master mix was prepared, consisting of: 1µL nuclease-free water, 12.5µL 2X buffer, 0.5µL reverse transcriptase/Taq polymerase enzyme, 2µL forward primer (10µM), 2µL reverse primers (10µM), and 2µL probe (2.5µM). To this mix, 5µL of extracted RNA was added per sample or control (positive and negative). The primer and probe sequences used were: GAC CRA TCC TGT CAC CTC TGA C (sense primer), AGG GCA TTY TGG ACA AAK CGT CTA (anti-sense primer) and TGC AGT CCT CGC TCA CTG GGC AGC (probe).

Meteorological data

From January 2017 to December 2022, twelve climatological parameters were retrieved from National Aeronautic and Space Administration (NASA) POWER Data Access Viewer via the link: https://power.larc.nasa.gov/data-access-viewer. These parameters included: earth surface temperature (EaST), temperature at 2 meters (T_2M), wet bulb temperature at 2 meters (WbT_2M), specific humidity at 2 meters (SH_2M), relative humidity at 2 meters (RH_2M), dew/frost point at 2 meters (DFPt_2M), average corrected precipitation (AvCP), cumulative rainfall (CumRF), solar radiation (SolRad), Maximum temperature at 2 meters (Tmax_2M), minimum temperature at 2 meters (Tmin_2M) and temperature range at 2 meters (Trange_2M). Meteorological data were obtained from ground stations located at latitude 3°9'N, longitude 11°5'E for the Center region; latitude 4°0'N, longitude 9°7'E for the Littoral region; latitude 5°3'N, longitude 10°4'E for the West region; and latitude 9°3'N, longitude 13°5'E for the North region.

Time-Series Analysis:

Monthly influenza-positive case data from 2017 to 2022 were obtained from the National Influenza Centre. These were treated as dependent variables, while the twelve monthly climate parameters served as independent variables. Two models were applied: Autoregressive Integrated Moving average (ARIMA), to assess the relationship between each climate variable and influenza activity; and the simple seasonal exponential smoothing to evaluate the influence of seasonal variations. The dataset was split into two chronological series: calibration phase from January 2017 to June 2022 (66 months), and validation phase from July to December 2022 (6 months). Model performance was assessed using stationary R² and root mean square error (RMSE). The ARIMA model (𝑝, d, 𝑞) (𝑃, 𝐷, 𝑄, 𝑄)s includes (a) auto-regression (AR), (b) moving average (MA) and (c) differencing, to achieve stationarity. The general ARIMA equation is:

𝑌_𝑡 =µ₁𝑌_𝑡−1 +µ₂𝑌_𝑡−2 +⋅⋅⋅+µ_𝑝𝑌_𝑡−𝑝 +𝑒𝑡 -µ₁𝑒_𝑡−1 -⋅⋅⋅ -µ_𝑞𝑒_𝑡−𝑞. ……………. (equation 1)

For seasonal smoothing, the model equations are:

S_t = α(y_t – I_t−L) + (1-α)(S_t−1 + b_t−1)……………. (Eq. 2)

b_t = γ(S_t – S_t−1) + (1 – γ)b_t−1……………. (Eq. 3)

I_t = β(y_t – S_t−1 – b_t−1) + (1 – β)I_t−L……………. (Eq. 4)

F_t+m = S_t +mb_t + I_t−L+m……………. (Eq. 5)

Where S_t is the level smoothing, b_t is the trend smoothing, I_t is the seasonal smoothing, F_t+m is the forecast at m periods ahead, L is the season length, y_t denotes the observation and t denotes the time. Finally, α, β, and γ are constants that take a value between 0 and 1.

Model performance of the ARIMA and the seasonal smoothing models were evaluated using root mean square error (RMSE) and stationary R².[18] SPSS version 23.0 software was used for all analysis and 𝑃 < 0.05 was considered statistically significant.

I- Description of the socio-demographic characteristics of the study population

The study population was predominantly composed of participants from the Centre region, followed by the North and West regions. Most enrolments occurred in 2019 for the Centre and West regions, and in 2021 for the Littoral and North regions. Age distribution revealed that the 0–2 and 2–4 year age groups were the most represented, whereas individuals over 50 years were the least represented. Female participants slightly outnumbered males in three of the four regions, in the exception of the Littoral region, where males were predominant. Influenza positivity rates varied by region, ranging from 20.0% in the West, 20.1% in the North, 33.0% in the Centre, to 39.3% in the Littoral region. Across all regions, influenza A was the predominant type, accounting for 70.0%-81.4% of confirmed cases. The socio-demographic characteristics of the study population are summarized in Table 1.

Table 1
Description of study population
Characteristic	Sub-categories	Regions				Total N (%)
Characteristic	Sub-categories	Centre N (%)	Littoral N (%)	North N (%)	West N (%)	Total N (%)
Year	2017	281 (17.4)	73 (12.7)	45 (3.5)	77 (9.9)	476 (11.2)
	2018	186 (11.5)	61 (10.6)	198 (15.3)	77 (9.9)	522 (12.2)
	2019	357 (22.1)	109 (19.0)	242 (18.7)	196 (25.1)	904 (21.2)
	2020	232 (14.4)	45 (7.8)	209 (16.2)	103 (13.2)	589 (13.8)
	2021	289 (17.9)	151 (26.3)	339 (26.2)	190 (24.4)	969 (22.7)
	2022	270 (16.7)	136 (23.7)	260 (20.1)	137 (17.6)	803 (18.8)
Age	0–2	512 (31.7)	309 (53.7)	320 (24.7)	237 (30.4)	1378 (32.3)
	2–4	311 (19.3)	158 (27.5)	371 (28.7)	142 (18.2)	982 (23.0)
	5–14	181 (11.2)	52 (9.0)	212 (16.4)	84 (10.8)	529 (12.4)
	15–49	264 (16.3)	26 (4.5)	164 (12.7)	164 (21.0)	618 (14.5)
	Above 50	84 (5.2)	2 (0.3)	18 (1.4)	132 (16.9)	236 (5.5)
	Missing	263 (16.3)	28 (4.9)	208 (16.1)	21 (2.7)	520 (12.2)
Sex	Male	663 (41.1)	303 (52.7)	555 (42.9)	346 (44.4)	1867 (43.8)
	Female	685 (42.4)	237 (41.2)	556 (43.0)	406 (52.1)	1884 (44.2)
	Missing	267 (16.5)	35 (6.1)	182 (14.1)	28 (3.6)	400 (9.4)
Influenza results	Positive	533 (33.0)	226 (39.3)	260 (20.1)	156 (20.0)	1175 (27.6)
Influenza results	Negative	1082 (67.0)	349 (60.7)	1033 (79.9)	624 (80.0)	3088 (72.4)
Influenza types	Influenza A	429 (80.5)	183 (81.0)	182 (70.0)	127 (81.4)	921 (78.4)
	Influenza B	96 (18.0)	37 (16.4)	78 (30.0)	26 (16.7)	237 (20.2)
	Influenza A&B	8 (1.5)	6 (2.6)	0	3 (1.9)	17 (1.4)
TOTAL		1615	575	1293	780	4263
* The most dominant categories by region are highlighted in bold.

II- Description of influenza positivity rates by region

Influenza virus was detected throughout the year in all regions, although its circulation patterns varied by year and location. Between 2017 and 2019, a bimodal distribution was observed in most regions, with a major peak between September and December and a minor peak between January and April. In 2017, the minor peak was absent in the Centre and Littoral regions. In certain years, the major peak occurred earlier, such as in July or August, and the timing of the minor peak shifted across regions. For example, in 2018, low-level circulation of influenza persisted from January to July, with region-specific peaks, while in 2019, the Centre region experienced a minor peak between April and June. In 2020, influenza activity was markedly reduced with little to no virus detection between January and July. The years 2021 and 2022 showed a further decline in influenza circulation, with no consistent seasonal pattern observed. These trends are illustrated in Fig. 2.

III- Description of meteorological parameters by region

We provide here a synthesized overview of regional climate profiles and their relevance to influenza dynamics. Overall, the climatological profiles across Cameroon’s four regions revealed distinct environmental signatures with implications for influenza transmission. The North region was characterized by high temperatures, low humidity, and minimal rainfall, while the Littoral region exhibited consistently high humidity, abundant rainfall, and moderate temperatures. Seasonal patterns were evident for most parameters: temperature indicators (T2M, EaST, and WbT2M) peaked between January and April; rainfall (AvCP, CumRF) intensified from July to October; and solar radiation (SolRad) showed troughs in January, particularly in the West and North regions. These climatic contrasts help explain regional differences in influenza activity and model performance. They also underscore the importance of integrating localized climate data into predictive frameworks, reinforcing the need for region-specific modeling approaches to improve influenza surveillance and early warning systems. These climatological trends are depicted in Fig. 3 and detailed in Supplementary Table S1.

Table S1: Characteristics of the study variables.

Description of ARIMA forecasting models for influenza activity

In the Centre region, three climatological variables: minimum temperature at 2 meters (Tmin2M), average corrected precipitation (AvCP) and cumulative rainfall (CumRF), were significantly associated with influenza activity (p < 0.005), explaining 9.6%, 30.0% and 29.5% of the observed variation, respectively, (Table 2). However, the ARIMA model incorporating all climate variables performed poorly, accounting for only 19.1% of influenza activity and yielding a relatively high RMSE of 20.203. As shown in Fig. 4, the model did not fit well during the calibration period (January 2017 to June 2022), and failed to produce accurate forecasts during the final six months.

Table 2
Summary of ARIMA and Seasonal models performance in the Centre region.
Models		Fit Models parameters
Models	RMSE		Stat. R²	Detail	Climate variable	Estimate	P-value
ARIMA (0,0,0)(0,0,0)	20.149	0,028		Lag 0	EaST	-3.048	0.175
ARIMA (0,0,0)(0,0,0)	20.260	0,018		Lag 0	T_2M	-2.733	0.287
ARIMA (0,0,0)(0,0,0)	20.362	0,008		Lag 0	WbT_2M	1.960	0.481
ARIMA (0,0,0)(0,0,0)	19.403	0.096		Lag 0	Tmin_2M	4.834	0.011*
ARIMA (0,0,1)(0,0,0)	10.526	0.300		Lag 0	AvCP	-1.113	0.026*
ARIMA (0,0,0)(0,0,0)	17.509	0.295		Lag 1	CumRF	0.856	< 0.001*
ARIMA (0,0,0)(0,0,0)	20.234	0.020		Lag 0	SolRad	0.790	0.254
ARIMA (0,0,0)(0,0,0)	20,217	0.022		Lag 0	RH_2M	0.497	0.236
ARIMA (0,0,0)(0,0,0)	20.351	0,009		Lag 0	DFPt_2M	1.934	0.452
ARIMA (0,0,0)(0,0,0)	19.940	0,049		Lag 0	Trange_2M	-1.838	0.075
ARIMA (0,0,0)(0,0,0)	20.362	0.008		Lag 0	SH_2M	1.960	0.481
ARIMA (0,0,0)(0,0,0)	20.400	0.004		Lag 0	Tmax_2M	1.007	0.607
ARIMA (0,0,0)(0,0,0)	20.203	0.191		Lag 0	ALL	-554.102	0.655
Seasonal Model	16.06	0.82		Lag 0	None	NA	0.852
Fit: fitting results, RMSE: Root mean square error, Stat.R²: Stationary R²

In the Littoral region, five climate variables: showed a significant association with influenza activity: earth surface temperature (EaST), temperature at 2 meters (T2M), wet bulb temperature at 2 meters (WbT2M), specific humidity at 2 meters (SH2M), and dew/frost point temperature at 2 meters (DFPt2M), were significantly associated with influenza activity, explaining between 18.0% and 30.8% of the observed variation (Table 3). The ARIMA model including all twelve climate variables yielded the best performance in this region, with the lowest RMSE of 6.149 and accounted for 21.1% of the observed influenza activity (Table 3). Although the model performed moderately well during the estimation period, it did not provide reliable predictions of influenza activity during the forecast period (Fig. 4).

Table 3
Summary of ARIMA and Seasonal models performance in the Littoral Region.
Models		Fit Models parameters
Models	RMSE		Stat. R²	Detail	Climate variable	Estimate	P-value
ARIMA(0,0,1)(0,0,0)	5.422	0,308		Lag 0	EaST	-3.900	0.001*
ARIMA (0,0,0)(0,0,0)	5.850	0.180		Lag 0	T_2M	-3.162	0.001*
ARIMA (0,0,0)(0,0,0)	4.989	0.230		Lag 0	WbT_2M	-4.683	0.001*
ARIMA (0,0,0)(0,0,0)	6.270	0.009		Lag 0	Tmin_2M	0.544	0.440
ARIMA (0,0,0)(0,0,0)	6.247	0,017		Lag 0	AvCP	0.161	0.301
ARIMA (0,0,0)(0,0,0)	6.277	0.007		Lag 0	CumRF	0,004	0.498
ARIMA (0,0,0)(0,0,0)	6.196	0.033		Lag 0	SolRad	0.306	0.147
ARIMA (0,0,0)(0,0,0)	6.231	0.022		Lag 0	RH_2M	0.289	0.237
ARIMA (0,0,1)(0,0,0)	5.649	0.249		Lag 0	DFPt_2M	-3.594	0.017*
ARIMA (0,0,0)(0,0,0)	6.298	0,000		Lag 0	Trange_2M	0.080	0.861
ARIMA (0,0,1)(0,0,0)	5.641	0.251		Lag 0	SH_2M	-3.279	0.015
ARIMA (0,0,0)(0,0,0)	6.269	0.010		Lag 0	Tmax_2M	0.413	0.434
ARIMA (0,0,0)(0,0,0)	6.149	0.211		Lag 0	All	-1085.54	0.267
Seasonal Model	5.530	0.75		Lag 0	None	NA	0.224
Fit: fitting results, RMSE: Root mean square error, Stat.R²: Stationary R²

In the North region, four climate variables—solar radiation (SolRad), Tmin2M, CumRF, and AvCP were significantly associated with influenza activity, explaining 15.0%, 27.8%, 33.0%, and 34.7% of the variation, respectively (Table 4). The ARIMA model including all climate variables accounted for 27.3% of influenza activity, based on stationary R². This model was able to identify influenza peaks during the forecast period, although with a lag of approximately two to three months (Fig. 4).

Table 4
Summary of ARIMA and Seasonal models performance in the North Region.
Models		Fit Models parameters
Models	RMSE		Stat. R²	Detail	Climate variable	Estimate	P-value
ARIMA (0,0,0)(0,0,0)	11.873	0.002		Lag 0	EaST	-0.171	0.701
ARIMA (0,0,0)(0,0,0)	11.885	0,000		Lag 0	T_2M	-0.081	0.886
ARIMA (0,0,0)(0,0,0)	11.826	0.010		Lag 0	WbT_2M	-0.257	0.420
ARIMA (0,0,0)(0,0,0)	11.296	0.278		Lag 1	Tmin_2M	-3.147	0.018*
ARIMA (0,0,0)(0,0,0)	10,299	0,347		Lag 1	AvCP	-2.518	< 0.001*
ARIMA (0,0,0)(0,0,0)	10.433	0.330		Lag 0	CumRF	-0.095	< 0.001*
ARIMA (0,0,0)(0,0,0)	12.062	0.150		Lag 0	SolRad	-0.173	0.006*
ARIMA (0,0,0)(0,0,0)	11.828	0.010		Lag 0	RH_2M	-0.043	0.425
ARIMA (0,0,0)(0,0,0)	11.838	0,008		Lag 0	DFPt_2M	-0.109	0.469
ARIMA (0,0,0)(0,0,0)	11.766	0,020		Lag 0	Trange_2M	0.338	0.255
ARIMA (0,0,0)(0,0,0)	11.798	0.015		Lag 0	SH_2M (g/kg)	-0.249	0.329
ARIMA (0,0,0)(0,0,0)	11.881	0.001		Lag 0	Tmax_2M (°C	0.091	0.793
ARIMA (0,0,0)(0,0,0)	11.138	0.273		Lag 0	All	-81.689	0.349
Seasonal Model	10.20	0.76		Lag 0	None	NA	0.944
Fit: fitting results, RMSE: Root mean square error, Stat.R²: Stationary R²

In the West region, influenza activity showed significant association with five climate variables: EaST (6.6%), DFPt_2M (21.5%), T_2M (43.5%), CumRF (38.5%) and SolRad (40.3%) (Table 5). The ARIMA model incorporating all climate variables explained 23.5% of influenza activity and yielded a relatively low RMSE (4.017). Despite this, the model did not adequately capture the observed peaks and performed only moderately well during the prediction period (Fig. 4).

Table 5
Summary of ARIMA and Seasonal models performance in the West region.
Models		Fit Models parameters
Models	RMSE		Stat. R²	Detail	Climate variable	Estimate	P-value
ARIMA (0,0,0)(0,0,0)	4.039	0.066		Lag 0	EaST (°C)	-1.142	0.037*
ARIMA (0,0,1)(0,0,0)	3.187	0.435		Lag 0	T_2M	-2.330	0.008*
ARIMA (0,0,0)(0,0,0)	4.180	0.000		Lag 0	WbT_2M (°C)	-0.045	0.918
ARIMA (0,0,0)(0,0,0)	4.176	0.002		Lag 0	Tmin_2M (°C)	-0.143	0.703
ARIMA (0,0,0)(0,0,0)	4.180	0.000		Lag 0	AvCP (mm)	-0.014	0.879
ARIMA (0,0,1)(0,0,0)	3.582	0.385		Lag 0	CumRF	0.017	< 0.001*
ARIMA (0,0,1)(0,0,0)	3.096	0.403		Lag 0	SolRad	-0.150	< 0.001*
ARIMA (0,0,0)(0,0,0)	4.157	0.011		Lag 0	RH_2M (%)	0.042	0.398
ARIMA (0,0,1)(0,0,0)	3.958	0.215		Lag 0	DFPt_2M	0.293	0.001*
ARIMA (0,0,0)(0,0,0)	4.147	0,016		Lag 0	Trange_2M (°C)	-0.201	0.316
ARIMA (0,0,0)(0,0,0)	4.174	0.003		Lag 0	SH_2M (g/kg)	0.703	0.670
ARIMA (0,0,0)(0,0,0)	4.091	0.043		Lag 0	Tmax_2M (°C	-0.452	0.097
ARIMA (0,0,0)(0,0,0)	4.017	0.235		Lag 0	All	-196.237	0.103
Seasonal Model	3.188	0.66		Level	Level	NA	0.005*
Fit: fitting results, RMSE: Root mean square error, Stat.R²: Stationary R²,

Seasonal prediction models for influenza activity

In contrast to the ARIMA models, the seasonal prediction models demonstrated superior performance across all regions. In the Centre region, the seasonal model explained 82% of influenza activity and provided a good fit throughout the study period (January 2017 to December 2022), with a lower RMSE of 16.067 (Table 6). However, no statistically significant association was found between influenza activity and seasonal variation (P = 0.852) (Fig. 5).

In the Littoral region, the seasonal model explained 75% of the influenza activity and produced accurate forecasts from July to December 2022. The model yielded a low RMSE (5.530), indicating better performance than the ARIMA model. Nonetheless, the association between influenza activity and seasonal variation was not statistically significant (P = 0.224).

In the North region, the seasonal model was the best-performing approach, explaining 76% of influenza activity and providing reliable predictions from June 2017 to December 2022. The model yielded a lower RMSE (10.20) compared to the ARIMA model, although no significant association was found between influenza activity and seasonal variation (P = 0.944).

In the Western region, the seasonal model explained 66% of influenza activity and produced accurate forecasts from June 2017 to December 2022. The model yielded a low RMSE (3.188) and demonstrated a statistically significant association between influenza activity and seasonal variation (P = 0.005), confirming its predictive accuracy in this region.

In summary, the comparative evaluation of ARIMA and seasonal models across Cameroon’s climatic zones revealed important regional differences in model performance and predictive accuracy. While ARIMA models identified significant associations between specific climate variables and influenza activity, their overall explanatory power remained modest, and predictive reliability was limited particularly during the forecast period. In contrast, seasonal models consistently outperformed ARIMA across all regions, yielding higher stationary R² values and lower RMSE scores. Notably, only the West region demonstrated a statistically significant association between influenza activity and seasonal variation. These findings suggest that while climate variables contribute meaningfully to influenza dynamics, their influence may be region-specific and better captured through tailored seasonal modeling approaches. This underscores the need for localized forecasting frameworks that integrate climatic variability to enhance influenza surveillance and early warning systems.

This study aimed to compare two forecasting approaches for influenza activity across four climatically distinct regions of Cameroon: a time series model incorporating twelve meteorological variables, and a simple seasonal model. The regions: Centre (grasslands), Littoral (equatorial forest), North (Sudano-Sahelian zone), and West (western highlands), exhibit diverse climatic profiles. Analysis of surveillance data from 2017 to 2022 revealed year-round circulation of influenza virus in all regions, with notable heterogeneity in seasonal patterns. Typically, a minor peak occurred between January and April, and a major peak between July and December, although the timing and presence of these peaks varied by year and region.

These findings align with previous studies in tropical and subtropical settings. For instance, Lau et al. reported a bimodal influenza pattern in Hangzhou, China, with a dominant winter peak and an occasional summer peak.[19] Similarly, Munshili et al. observed dual peaks in the North region of Cameroon,[20] while studies by Heraud et al., Monamele et al., and N'gattia and al., confirmed continuous influenza circulation in tropical climates.[19, 21] Such variability suggests that climatic and seasonal factors may modulate influenza transmission differently across ecological zones.

In our analysis, the influence of meteorological variables on influenza activity varied by region. In the Centre region, rainfall accounted for approximately 30% of influenza cases, consistent with earlier findings from 2009–2015, which attributed 29.5% of cases to rainfall.[21] Rainfall may facilitate transmission by increasing indoor crowding and enhancing fomite-mediated spread.[22] In the North region, rainfall explained over 33% of influenza activity, contrasting with earlier findings that emphasized relative humidity as the dominant factor.[11] This discrepancy may stem from differences in dataset duration and climatic variability, as weather patterns become increasingly unpredictable due to climate change.[23]

In the Littoral and West regions, temperature-related variables—earth surface temperature (EaST) and temperature at 2 meters (T2M)—were most strongly associated with influenza activity, explaining up to 30.8% and 43.5% of cases, respectively. These results are consistent with studies from Wuhan, China, and temperate regions, where extreme temperatures have been linked to increased influenza transmission.[24] Tamerius et al. demonstrated that cold temperatures are key predictors of seasonal influenza in high latitudes, a finding supported by Lowen et al., who observed up to 75–100% transmission efficiency at 5.8°C in experimental settings.

Weaker associations were observed with humidity and solar radiation. In the Littoral region, specific humidity was linked to 25.1% of influenza cases, while solar radiation was associated with influenza activity in the North (15.0%) and West (40.3%) regions. In similar studies conducted in Central America and South Asia, specific humidity and sunshine duration have shown variable associations with influenza risk.^8,15,25 For example, increased sunshine was associated with a 37% reduction in influenza A risk for each hour of increased sunshine.[25]

Despite these associations, the ARIMA models demonstrated limited predictive power, with lower stationary R² values and inconsistent performance across regions. In contrast, the simple seasonal models, independent of climate variables, provided superior fits, explaining between 66% and 82% of influenza activity. These results suggest that influenza seasonality in Cameroon may be influenced more by socio-behavioral factors than by meteorological conditions alone. Previous studies have noted that influenza patterns in Cameroon often resemble northern hemisphere seasonality (October to March), with occasional southern hemisphere-like activity (April to September).[18] Biennial peaks have also been reported, typically between September and December.[18, 26] However, our data did not confirm consistent seasonal patterns in any of the regions, highlighting the complexity and variability of influenza dynamics in tropical settings.

Several limitations should be acknowledged in this study. First, the use of national surveillance data may have introduced bias due to inconsistent sampling throughout the year, potentially leading to under- or overestimation of influenza activity during certain periods. Second, the study did not account for socio-behavioural determinants of transmission, such as holiday periods, access to health services, income, education level, race, ethnicity, and COVID-19 related restrictions, all of which may influence influenza dynamics independently of climate.

This study demonstrated that the seasonal model outperformed the ARIMA model in describing and predicting influenza activity in four climatically distinct regions of Cameroon. The seasonal model showed superior fit and predictive accuracy, particularly in the West region, where its performance was supported by statistically significant association, lower RMSE and higher stationary R² values. These findings suggest that simple seasonal models may offer reliable and timely forecasts of influenza epidemics, especially in resource-limited settings. By improving short-term prediction capacity, such models can support more effective prevention strategies, optimize the allocation of health resources, and ultimately contribute to reducing the burden of influenza in Cameroon.

Data availability statement: The authors confirm that the data supporting the findings of this study are available within the article.

Ethical statement

The samples analyzed in this study were collected as part of the national influenza surveillance program coordinated by the Department of Epidemic and Pandemic Control of the Ministry of Public Health, Cameroon. Ethical approval N°3971CEI-Udo/07/2023/M was required for this study. All individual data were anonymized with a code and were entered in a password protected database.

Acknowledgments

We would like to thank the regional coordinators, and the various focal points involved in sentinel influenza surveillance in Cameroon.

Financial support

This research was funded by the U.S. Department of Health and Human Services, DHHS (grant number 6 DESP060001-01-01) and the Centers for Disease Control and Prevention (CDC Atlanta), through HEADA, Inc. Contract N° 2022-01-CDC Flu Surv.

Author Contributions

Conceptualization: Gwladys C. Monamele, Richard Njouom.

Formal analysis: Desmon Toutou Tsafack, Chavely Gwladys Monamele, Mohamadou Ripa Njankouo, Christian Nsangou, Hermann Landry Munshili Njifon.

Funding acquisition: Richard Njouom, Ubald Tamoufe.

Methodology: Desmon Toutou Tsafack, Chavely Gwladys Monamele, Mohamadou Ripa Njankouo, Hermann Landry Munshili Njifon, Francioli Koro Koro

Software: Desmon Toutou Tsafack, Chavely Gwladys Monamele, Mohamadou Ripa Njankouo, Francioli Koro Koro

Supervision: Chanceline Ndongo Bilounga, Ubald Tamoufe, Ronald Perraut, Francioli Koro Koro, Richard Njouom

Validation: Chanceline Ndongo Bilounga, Ubald Tamoufe, Ronald Perraut, Francioli Koro Koro, Richard Njouom,

Writing –original draft: Desmon Toutou Tsafack, Chavely Gwladys Monamele.

Writing –review & editing: Desmon Toutou Tsafack, Chavely Gwladys Monamele, Francioli Koro Koro, Mohamadou Ripa Njankouo, Hermann Landry Munshili Njifon, Chanceline Ndongo Bilounga, Ubald Tamoufe, Ronald Perraut, Richard Njouom.

Conflict of Interest Statement

The authors declare that no conflict of interest exists.

WHO launches new global influenza strategy. [cited 6 Mar 2024]. Available: https://www.who.int/news/item/11-03-2019-who-launches-new-global-influenza-strategy
Touré CT, Fall A, Andriamandimby SF, Jallow MM, Goudiaby D, Kiori D, et al. Epidemiology and Molecular Analyses of Influenza B Viruses in Senegal from 2010 to 2019. Viruses. 2022;14: 1063. doi:10.3390/v14051063
Influenza. [cited 20 Mar 2024]. Available: https://www.who.int/teams/health-product-policy-and-standards/standards-and-specifications/vaccine-standardization/influenza
Brian S Finkelman, Cecile V, Mattheu J Ferrari. Global patterns in seasonal activity of influenza A/H3N2, A/H1N1, and B from 1997 to 2005: viral coexistence and latitudinal gradients - PubMed. Dec 2007 [cited 6 Feb 2024]. Available: https://pubmed.ncbi.nlm.nih.gov/18074020/
Cecile V. Influenza epidemics in the United States, France, and Australia, 1972-1997 - PubMed. Jan 2004 [cited 6 Feb 2024]. Available: https://pubmed.ncbi.nlm.nih.gov/15078594/
Azziz Baumgartner E, Dao CN, Nasreen S, Bhuiyan MU, Mah-E-Muneer S, Al Mamun A, et al. Seasonality, timing, and climate drivers of influenza activity worldwide. J Infect Dis. 2012;206: 838–846. doi:10.1093/infdis/jis467
Cecile V, Wladimir J Alonso, Lone Simonsen. Influenza in tropical regions - PubMed. Apr 2006 [cited 6 Feb 2024]. Available: https://pubmed.ncbi.nlm.nih.gov/16509764/
Soebiyanto RP, Clara W, Jara J, Castillo L, Sorto OR, Marinero S, et al. The Role of Temperature and Humidity on Seasonal Influenza in Tropical Areas: Guatemala, El Salvador and Panama, 2008–2013. Colijn C, editor. PLoS ONE. 2014;9: e100659. doi:10.1371/journal.pone.0100659
Tamerius J, Nelson MI, Zhou SZ, Viboud C, Miller MA, Alonso WJ. Global influenza seasonality: reconciling patterns across temperate and tropical regions. Environ Health Perspect. 2011;119: 439–445. doi:10.1289/ehp.1002383
Monamele GC, Vernet M-A, Nsaibirni RFJ, Bigna JJR, Kenmoe S, Njankouo MR, et al. Associations between meteorological parameters and influenza activity in a subtropical country: Case of five sentinel sites in Yaoundé-Cameroon. Shaman J, editor. PLOS ONE. 2017;12: e0186914. doi:10.1371/journal.pone.0186914
Munshili Njifon HL, Monamele CG, Kengne Nde C, Vernet M, Bouba G, Tchatchouang S, et al. Influence of meteorological parameters in the seasonality of influenza viruses circulating in Northern Cameroon. Influenza Other Respir Viruses. 2019;13: 158–165. doi:10.1111/irv.12612
Nisar N, Badar N, Aamir UB, Yaqoob A, Tripathy JP, Laxmeshwar C, et al. Seasonality of influenza and its association with meteorological parameters in two cities of Pakistan: A time series analysis. PLoS ONE. 2019;14: e0219376. doi:10.1371/journal.pone.0219376
Wang J, Zhang L, Lei R, Li P, Li S. Effects and Interaction of Meteorological Parameters on Influenza Incidence During 2010–2019 in Lanzhou, China. Front Public Health. 2022;10: 833710. doi:10.3389/fpubh.2022.833710
N’gattia AK, Coulibaly D, Nzussouo NT, Kadjo HA, Chérif D, Traoré Y, et al. Effects of climatological parameters in modeling and forecasting seasonal influenza transmission in Abidjan, Cote d’Ivoire. BMC Public Health. 2016;16: 972. doi:10.1186/s12889-016-3503-1
Soebiyanto RP, Clara WA, Jara J, Balmaseda A, Lara J, Lopez Moya M, et al. Associations between seasonal influenza and meteorological parameters in Costa Rica, Honduras and Nicaragua. Geospatial Health. 2015;10. doi:10.4081/gh.2015.372
Monamele GC, Tsafack DT, Bilounga CN, Njankouo Ripa M, Nsangou Yogne C, Munshili Njifon HL, et al. The Detection of Influenza Virus Before and During the COVID-19 Pandemic in Cameroon. Influenza Other Respir Viruses. 2024;18: e13313. doi:10.1111/irv.13313
World Bank Group. World Bank Climate Change Knowledge Portal. 2021 [cited 21 Nov 2024]. Available: https://climateknowledgeportal.worldbank.org/
Chadsuthi S, Iamsirithaworn S, Triampo W, Modchang C. Modeling Seasonal Influenza Transmission and Its Association with Climate Factors in Thailand Using Time-Series and ARIMAX Analyses. Comput Math Methods Med. 2015;2015: 436495. doi:10.1155/2015/436495
Lau SY, Cheng W, Yu Z, Mohammad KN, Wang MH, Zee BC, et al. Independent association between meteorological factors, PM2.5, and seasonal influenza activity in Hangzhou, Zhejiang province, China. Influenza Other Respir Viruses. 2021;15: 513–520. doi:10.1111/irv.12829
Munshili Njifon HL, Monamele CG, Kengne Nde C, Vernet M-A, Bouba G, Tchatchouang S, et al. Influence of meteorological parameters in the seasonality of influenza viruses circulating in Northern Cameroon. Influenza Other Respir Viruses. 2019;13: 158–165. doi:10.1111/irv.12612
Monamele GC, Vernet M-A, Nsaibirni RFJ, Bigna JJR, Kenmoe S, Njankouo MR, et al. Associations between meteorological parameters and influenza activity in a subtropical country: Case of five sentinel sites in Yaoundé-Cameroon. PloS One. 2017;12: e0186914. doi:10.1371/journal.pone.0186914
Tamerius JD, Shaman J, Alonso WJ, Bloom-Feshbach K, Uejio CK, Comrie A, et al. Environmental predictors of seasonal influenza epidemics across temperate and tropical climates. PLoS Pathog. 2013;9: e1003194. doi:10.1371/journal.ppat.1003194
Lane MA, Walawender M, Carter J, Brownsword EA, Landay T, Gillespie TR, et al. Climate change and influenza: A scoping review. J Clim Change Health. 2022;5: 100084. doi:10.1016/j.joclim.2021.100084
Li Y, Wu J, Hao J, Dou Q, Xiang H, Liu S. Short-term impact of ambient temperature on the incidence of influenza in Wuhan, China. Environ Sci Pollut Res. 2022;29: 18116–18125. doi:10.1007/s11356-021-16948-y
Imai C, Brooks WA, Chung Y, Goswami D, Anjali BA, Dewan A, et al. Tropical influenza and weather variability among children in an urban low-income population in Bangladesh. Glob Health Action. 2014;7: 24413. doi:10.3402/gha.v7.24413
Igboh L, Roguski K, Marcenac P, Emukule G, Tempia S, Al E et. Timing of seasonal influenza epidemics for 25 countries in Africa during 201019 a retrospective analysis. 2024 [cited 12 Dec 2024]. Available: https://hdl.handle.net/10539/39378

SupplementaryTableS1.docx

Download PDF

Reviewers agreed at journal
17 Nov, 2025
Reviewers invited by journal
23 Oct, 2025
Editor assigned by journal
18 Oct, 2025
First submitted to journal
17 Oct, 2025

You are reading this latest preprint version

Evaluation of time series models for influenza trends accross climatic zones in Cameroon (2017-2022)

Status:

Version 1

Abstract

Background:

Methods:

Results:

Conclusions:

Figures

Introduction

Material and methods

Data sources and study design

Geographic and climatic context

Laboratory analyses

Meteorological data

Time-Series Analysis:

Results

I- Description of the socio-demographic characteristics of the study population

II- Description of influenza positivity rates by region

III- Description of meteorological parameters by region

Description of ARIMA forecasting models for influenza activity

Seasonal prediction models for influenza activity

Discussion

Conclusions

Declarations

References

Supplementary Files

Status:

Version 1